Abstract

The developed classical machine learning models based on linear models and decision trees, the modern algorithms of convolutional neural networks, and the neural network autoencoder are compared in solving the problem of predictive detection of pre-failure and failure states of aircraft engines. The NASA data set includes the sensor readings reflecting the life cycle of aircraft engines. Several problem formulations are investigated in the study: (i) the problem of binary and multiclass classification, where the normal, pre-failure, and failure states of aircraft engines are predicted; (ii) the regression problem intended to predict the accurate number of working cycles to engine failure; and (iii) the unsupervised learning, where the neural network autoencoder is applied to detect abnormal cycles of aircraft engine operation. The obtained algorithms are combined in a framework useful in analyzing a wide spectrum of data of predictive maintenance.

中文翻译：

---

预测飞机发动机故障前和故障状态的机器学习算法的比较分析

摘要

比较了基于线性模型和决策树开发的经典机器学习模型，卷积神经网络的现代算法以及神经网络自动编码器，以解决飞机发动机故障和故障状态的预测检测问题。NASA数据集包括反映飞机发动机生命周期的传感器读数。在研究中研究了几种问题公式：（i）二元和多类分类问题，其中预测了飞机发动机的正常，故障前和故障状态；（ii）回归问题旨在预测发动机故障的准确工作循环数；（iii）无监督学习，其中将神经网络自动编码器用于检测飞机发动机运行的异常周期。